In partnership with

ISSUE 0x0E · TL;DR 2026-10-19 · ~6 MIN
⚡ aiBytes_ #14

Hello, builders 👋

Welcome back to your weekly dose of AI Bytes. I keep coming back to one question this week: if a company can replace git in five months, what happens to the companies that sell software? 🌳 Anthropic built its own version control system, Scion, because its merges went from about 2,500 a day in May to over 30,000 in September.

Build is beating buy: Anthropic also built its own code review and code search in the same five months, one developer used Claude to make free alternatives to Word, Excel and PowerPoint, and Thorsten Ball is asking why we still open pull requests. Anthropic says most companies won't hit these limits soon. I don't think the software business is ending, but it has a new competitor: its own customers, building instead.

In this issue

🌍 AI News 9 stories
📣 Official Announcements Meta · OpenAI · Google
0x02 Viral on X 5 posts
0x03 New AI Products 5 launches
0x04 Trending Github Projects 4 repos
0x05 HN Deep Stories 4 threads
🌍 AI NEWS

🌳 Anthropic replaced git with its own version control in five months - Merges grew from about 2,500 a day in May to over 30,000 in September, so Claude helped build Scion, plus code review and code search. Story of the week (Anthropic)

🔐 Claude explains where the ECDSA curve seeds came from - Anthropic says Claude traced the origin of the seed behind NIST's P-256 curve, a mystery cryptographers argued over for decades. (Anthropic)

🧯 Experts blame the AI hacking scares on labs, not rogue agents - Containment and monitoring failures at frontier labs are behind the recent reports, and CISOs should vet their AI vendors. (CSO Online)

💔 Replit is no longer working with Anthropic - CEO Amjad Masad cites cost among several reasons; Replit announced an OpenAI partnership in August. (TechCrunch)

🪦 Character.AI files for bankruptcy - Five years in, the chatbot platform will sell some of its assets to Disney for $20 million. (TechCrunch)

📱 Google Play now runs private, vibe-coded Android apps - Build a personal app with AI and use it without publishing it to everyone on the Play Store. (TechCrunch)

💳 Amex sets rules for AI agents that shop - A framework meant to ensure agents only make the purchases their users have authorized. (TechCrunch)

⚖️ Z.ai releases GLM-Legal - A GLM model aimed at legal work. (Z.ai)

👁️ Elorian's first model reasons with images - By Elorian's own evals, Glimpse beats GPT-6 Astra on scientific chart reasoning and short video captioning. (TechCrunch)

"Writing code stopped being the slow part. What slows us down now is everything built around it for humans: the source control, the reviews, the containment."

📣 OFFICIAL ANNOUNCEMENTS

This week's launches, in the companies' own words.

Meta @AIatMeta · THU 15

Today we’re releasing Muse Spark Open, the most capable US open-weight model to date.

Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.

Key Model Specs:
→ 816B total / 42B active parameters
→ 1M token context window
→ Natively multimodal across text, image, video, audio

Post on X ↗ · research.meta.ai ↗

OpenAI @OpenAI · FRI 16

GPT-6 Astra Pro in ChatGPT got an upgrade with faster responses and improved use of plugins and tools.

And now, multiple agents can tackle different parts of a task in parallel. You can follow each agent’s progress and open completed tasks to review their work.

Now available to Pro users in the Chat tab.

Post on X ↗

Google @GoogleAI · TUE 13

Gemini 4️⃣ Argon ✨ is now rolling out globally.

Our frontier model is built for complex, real-world tasks across coding, finance, legal, cybersecurity, and science.

Check out how you can use Argon in your favorite Google product 🧵

Post on X ↗

0x02 VIRAL ON X
 

The AI posts that drew the biggest reaction this week.

🚀 @felixrieseberg Hi, I'm Felix. I'm an eng lead working on Claude. I love watching people take flight and do things they never thought possible. / We often …♥︎ 22.8k
🔐 @sweis CLAUDE FOUND THE SEEDS / https://www.anthropic.com/research/ecdsa-curve-origins♥︎ 9.3k
📄 @unusual_whales BREAKING: A developer used AI to build free, open-source alternatives to Microsoft Word, Excel and PowerPoint. / Brandon Thomas, creator of …♥︎ 6.3k
🔁 @vinhnx If you're building a coding agent harness, give your agent this paper and let it figure out how to improve itself. …♥︎ 3.8k
🔀 @thorstenball Why are you still using pull requests? / Is it for code review? / If so, do you really think your or your colleagues catch bugs and mistakes …♥︎ 1.6k
0x03 NEW AI PRODUCTS
 
🥞 Pancake - Pitched as the Vercel for GTM: deploy go-to-market ideas from Claude or Codex, with sending and 40+ data providers behind one MCP connection. (ProductHunt) ▲ 585
🔎 ReSO AI - Generative engine optimization: helps brands get recommended and cited in ChatGPT, Perplexity and Google AI Overviews. (ProductHunt) ▲ 509
🎞️ Jockey by TwelveLabs - Organizes and searches a whole photo and video library, from inside Claude or ChatGPT or through its API. (ProductHunt) ▲ 415
♻️ Inferock - Serverless pay-per-token inference that shows where calls fail and credits back eligible failed calls. (ProductHunt) ▲ 400
🗣️ Onepin - Checks AI voiceover for names, prices and dates, and fixes a wrong word without re-rendering the take. (ProductHunt) ▲ 384
0x04 TRENDING GITHUB PROJECTS
 
🔍 morluto/rea - Reverse engineer anything with agents, from app behavior down to native binaries. +248,962 ★
📐 cathrynlavery/diagram-design - Editorial diagrams for Claude Code, Codex, Copilot, Cursor and more: 44 types, as self-contained HTML and SVG. +9,024 ★
🗜️ mksglu/context-mode - Context-window optimization for coding agents: sandboxes tool output and keeps session memory, via MCP and hooks. +4,207 ★
🧩 anthropics/claude-plugins-official - Anthropic's official directory of Claude Code plugins. +2,328 ★
0x05 HN DEEP STORIES
 
🧠 Neural networks are already clean - Goodfire's working hypothesis about neurons, generalization and the ecology of circuits. (HackerNews) 807 pts
🔨 AI is a tool, and using it well takes skill - A reply to Rich Hickey's "'AI' is Not a Tool". (HackerNews) 802 pts
📏 The gap in shared understanding of LLM capability is widening - Andrej Karpathy's post, and the week's most-argued thread. (HackerNews) 727 pts · 1,072 💬
🗺️ 90% of US working hours are spent on tasks AI benchmarks barely cover - Mercor maps which parts of the labor force public agentic benchmarks actually measure. (HackerNews) 571 pts

🙏 Thanks to Wispr Flow for sponsoring

The AI notetaker that gets the hard words right

Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.

Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.

Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.

Thank you for reading. If you have any feedback, feel free to respond to this email. See you next week!